Swap any face in a video.
Give it a video and a photo of the face you want. It replaces one person, in every frame, and leaves everything else exactly as it was filmed.
Three inputs. One result.
No editing, no masks, no frame-by-frame work.
Upload the video
Any clip with a person in it. Up to 1080p, and the original sound is kept.
The video is analysed and every face is found and followed.Add the reference face
A clear, front-facing photo of the face you want in the video. The sharper and more frontal it is, the closer the match.
You confirm you have the right to use that face.Choose who, get the result
Click the person to replace. Only that person changes, and detail is restored and upscaled to the size you pick.
Several people in the clip? You choose; it never guesses.See what it does to real footage.
Every clip is a real render. Drag the divider: before on the left, after on the right. The small portrait is the reference face that was used.
Match = how close the result's face is to the reference photo, measured with the SFace face-recognition model (cosine similarity). The tick marks 0.36, the usual threshold for “same person”. Footage: Mixkit free stock; the faces are other stock models, used for testing only.
Multiple people. You decide who changes.
One real clip, three people. Click anyone: only that person is replaced, and the others, the street and the sound stay exactly as filmed. Each result was rendered from this same clip.
Built for the shots where other swaps fall apart.
Most tools swap whatever face they find. This one is built around choosing the person, following them through hard frames, and telling you the truth about what it did.
Swapped faces come out smooth. We make them vivid.
A swap on its own leaves faces over-smooth and a little waxy. Every result goes through a state-of-the-art diffusion model that rebuilds skin texture, eyes, teeth and hair edges, so the face looks alive instead of plastic. Then it is delivered at 1080p or 4K with the original sound.
Same frames, same swap. Left: the swap alone. Right: after diffusion face enhancement.Only the person you choose
Click a face, draw a box, match a photo, or take the left one. Everyone else, the background and the sound stay as filmed.
Two-person test: the person left alone stayed at 0.99 similarity to their own original.Survives head turns
Face swap eases back to the original when a head turns too far, instead of smearing. Head swap follows the head even when the face detector loses it.
Average share of the chosen person’s frames that were swapped, 74 public clips.Steady, never jumpy
Landmarks are smoothed without lag and swap strength follows the head angle frame by frame, so the new face does not flicker or hop.
Blank frames in 74 swapped clips. Frame count, rate and size always match the input.Looks like the reference
The result’s face moves from nothing like the reference to clearly like it. Front-facing references give the strongest match.
Mean similarity to the reference, 74 clips, against 0.08 before the swap (same person above 0.36).Vivid faces, not plastic ones
A diffusion model rebuilds skin, eyes and teeth after the swap, and delivers at 1080p or 4K with the original audio put back untouched.
Fast mode uses a lighter enhancer: seconds instead of minutes.Says no instead of guessing
Two equally prominent people: it shows the candidates and renders nothing. No usable face: it tells you why instead of producing a bad result.
About 1 in 20 public test clips had no usable face and was reported that way.The identity changes. The rest stays.
Keeps the original hair, head shape and expression. Best for dialogue, interviews and long clips.
- Replaces
- The face
- Keeps
- Hair, head shape, expression
- Speed
- Seconds for the swap, then restore and upscale
- Best for
- Most clips, up to minutes long
The whole head changes.
Hair, head shape and face all come from the reference. Best for new looks and short hero shots.
- Replaces
- Face, hair and head shape
- Keeps
- Body, clothes, background, everyone else
- Speed
- Minutes for a few seconds of video
- Best for
- Short clips, up to 12 s
What it needs, and what it won’t do.
Face swapping needs permission, and it works best with the right inputs.
For the best result
- A sharp, front-facing reference photo with the whole face visible
- A face that is large enough in the video to see clearly
- Faces that are not covered by hands, microphones or props
- Even light: in heavy lens flare or strong backlight the face can be lost for a moment, and those frames stay as filmed
- Short heads-up: head swap takes minutes, face swap takes seconds
What it will not do
- Start without confirming you have the right to use the reference face
- Guess when two people are equally prominent: you choose
- Swap faces turned beyond the usable angle: those frames stay original and are counted in the report
- Keep your files: uploads and results are deleted after 24 hours